The-Good-Guy-Daily

Sunday, November 22, 2026Economy · VFINX · VBMFX · Gold

Made automatically every morning by scripts on a home computer (the economy notes use a small local AI model and are fact-checked by script). Not financial advice.

US Economy — Sunday, November 22, 2026

Generated 2026-11-22 09:40 by Good-Guy-Economy. Facts: FRED, BLS, BEA, Fed, CBO feeds; headlines: AP and Reuters.

Morning note

What changed from official data

  • No new official data since the previous report.
  • New releases: Activities of U.S. Multinational Enterprises, 2024 (BEA, 11/20); Recent and Upcoming Presentations by the Staff of CBO's Health Analysis Division (CBO, 11/20); An Analysis of the Navy's Fiscal Year 2027 Shipbuilding Plan (CBO, 11/19).

What it means ANALYSIS written by script from the computed history
[ANALYSIS] None of the 8 recession warning lights is on; since 1986 a recession began within a year after only 2% of months like this, against 11% for any month. None of the conditions true now has come before recessions more often than usual. Direction of the latest readings: growth up, jobs flat, inflation up, rates flat, recession risk up.

Watch next official release calendars

  • GDP (Second Estimate) and Corporate Profits, 3rd Quarter 2026 — Wed Nov 25 (BEA)
  • Personal Income and Outlays, October 2026 — Wed Nov 25 (BEA)
  • JOLTS job openings for October 2026 — Tue Dec 1 (BLS)

✓ Number check: every figure in the note matches tonight's data.

Where it's heading (daily scorecard)

Where the economy is heading:

- Growth: up (medium confidence) - Real GDP growth (annualized) 3.4 % for 2026 Q3 (was 2.2)

- Jobs: flat (low confidence) - Initial jobless claims (weekly) 203,000 for 2026-11-14 (was 200,000)

- Inflation: up (high confidence) - CPI inflation y/y 3.72 % for 2026-09 (was 3.35)

- Rates: flat (low confidence) - Fed funds target (upper) 4 % for 2026-11-22 (was 4)

- Recession risk: up (medium confidence) - Yield curve 10y-3m 0.6 pt for 2026-11-20 (was 0.88)

- Scorecard so far: 2 right, 0 wrong, 1 no change, of 3 calls checked

Key numbers

IndicatorLatestPeriodPrevious~1 yr ago
Real GDP growth (annualized)3.4 %2026 Q3▲ 2.23.9
Unemployment rate4.1 %2026-10▼ 4.24.4
Payrolls, monthly change165 k jobs2026-10▲ 60-140
Initial jobless claims (weekly)203,000 2026-11-14▲ 202,000228,000
Sahm recession indicator0 pt2026-10■ 00.23
CPI inflation y/y3.72 %2026-09▲ 3.353.02
Core CPI y/y2.57 %2026-09▲ 2.453.02
Core PCE y/y3.12 %2026-09▲ 3.032.82
Fed funds target (upper)4 %2026-11-22■ 44NEW
Fed funds effective3.88 %2026-11-19■ 3.883.88
10-yr Treasury5.04 %2026-11-19▼ 5.074.13
Yield curve 10y-2y0.29 pt2026-11-20▲ 0.280.55
Yield curve 10y-3m0.6 pt2026-11-20▼ 0.70.16
30-yr mortgage rate7.28 %2026-11-19▲ 7.116.24
High-yield spread3.52 pt2026-11-19▲ 3.473.17
Corporate Baa spread1.56 pt2026-11-19▲ 1.521.79
Consumer sentiment (UMich)51.5 2026-11▲ 4851
Real retail sales y/y2.75 %2026-09▲ 2.290.6
Industrial production y/y2.15 %2026-10▲ 2.141.76
Housing starts (annual rate)1,309 k2026-10▲ 1,1911,273
WTI oil89.53 $/bbl2026-11-17▲ 89.2460.66

NEW = a value published since the previous run. Inflation, retail sales and industrial production are year-over-year %; payrolls are the monthly change.

Today vs history computed from FRED + NBER recession dates — no model involved

Recession warning lights: 0 of 8 on (2026-10)

Since 1986, in months with no lights on, a recession began within 12 months 2% of the time (4 of 241 months). Base rate for any month: 11%.

Conditions true now, and what followed each past time they began:

ConditionNowPast casesRecession ≤12 mo≤24 moUnemployment 12 mo later
Consumer sentiment below 6051.5 (2026-11)6 since 19520 of 30 of 2-0.2 pt
→ LOWER recession risk than usual (0 of 3 = 0% vs 17% for any month since the data begins); unemployment usually little changed afterwards
CPI inflation above 3%3.72 (2026-09)16 since 19484 of 165 of 15-0.3 pt
→ about the USUAL recession risk (4 of 16 = 25% vs 18% for any month since the data begins) - no extra signal; unemployment usually FELL afterwards
30-yr mortgage rate above 7%7.28 (2026-11-19)3 since 19710 of 30 of 3-0.2 pt
→ LOWER recession risk than usual (0 of 3 = 0% vs 14% for any month since the data begins); unemployment usually little changed afterwards

Not true now: Unemployment up 0.5 pt or more from its 12-month low; Yield curve inverted (10-yr below 3-month); Yield curve inverted (10-yr below 2-yr); Jobless claims 20%+ above their 12-month low; Corporate bond spread 1 pt+ above its 12-month low; Industrial production lower than a year ago; Housing starts 15%+ below a year ago; Job growth under 50k a month (3-month average); Oil price up 50%+ from a year ago; 10-yr Treasury yield up 1 pt+ from a year ago; Fed funds rate up 1 pt+ from a year ago; Fed funds rate down 1 pt+ from a year ago

Official releases (last 3 days)

Headlines (AP, Reuters — last 24 hours)

News notes summarised by the model from official text — sources named per item

No new data available

Source: N/A

  • [DATA] None. The Script Output indicates that "nothing new from official sources in the last 3 days" was found for the READING PACK section.
  • GAPS: Economic data for the last 3 days could not be fetched or is reported as unavailable.

    BEA: U.S. Multinational Enterprise Employment

    Source: https://www.bea.gov/news/2026/activities-us-multinational-enterprises-2024

  • [DATA] Worldwide employment by U.S. multinational enterprises increased 0.9 percent to 44.2 million workers in 2024 (preliminary) from 43.8 million workers in 2023 (revised), according to statistics released today by the U.S. _(corrected by fact check)_
  • [DATA] Employment in the United States by U.S. parents increased 0.1 percent to 29.9 million workers in 2024.
  • [DATA] U.S. parents accounted for 67.6 percent of worldwide employment by U.S. multinational enterprises (MNEs), down from 68.1 percent in 2023.
  • [DATA] Employment abroad by foreign affiliates of U.S. MNEs increased 2.7 percent to 14.3 million workers, accounting for 32.4 percent of employment by U.S. MNEs worldwide.
  • GAPS: Could not check data from CBO publications or official BLS releases as they were listed but the full text was not provided in the reading pack for summarization.

    Fed Funds Target Update

    Source: N/A (Script Output Data)

  • [DATA] The Federal Funds target (upper) is 4 % for November 22, 2026.
  • [ANALYSIS] This represents a maintenance of the current federal policy rate level based on the script output's new data point since the last run.
  • GAPS: None.

    CBO Health Analysis Division Presentations

    Source: https://www.cbo.gov/publication/62921

  • [REPORTED] Recent and upcoming presentations by the Staff of CBO's Health Analysis Division are available through this publication.
  • GAPS: Full data extraction requires reading the external document.

    Headline Check

    web_search{query: "Inflation, economic growth and the mood of US consumers in focus this week"}

    Fact check (model + script, 5 notes): 3 confirmed against the source, 1 replaced with the source's own sentence, 0 removed, 1 not checked.

    History comparisons computed by econ-history

    S-20261122-1: Consumer sentiment below 60
    HISTORY: 6 past episodes since 1952; recession within 12 months in 0 of 3, within 24 months in 0 of 2
    WHAT IT SUGGESTS: [ANALYSIS] In the past this condition was followed by a recession less often than usual, so it is not a warning sign on its own.
    S-20261122-2: CPI inflation above 3%
    HISTORY: 16 past episodes since 1948; recession within 12 months in 4 of 16, within 24 months in 5 of 15
    WHAT IT SUGGESTS: [ANALYSIS] In the past this condition made no real difference to recession odds, so it is not a warning sign on its own.
    S-20261122-3: 30-yr mortgage rate above 7%
    HISTORY: 3 past episodes since 1971; recession within 12 months in 0 of 3, within 24 months in 0 of 3
    WHAT IT SUGGESTS: [ANALYSIS] In the past this condition was followed by a recession less often than usual, so it is not a warning sign on its own.

    Open direction calls ANALYSIS by rules: the direction of the latest readings

    VFINX vs TRENDmade Sat Nov 21

    VFINX vs trend - 2026-11-20

    VFINX closed at $638.66 on Fri Nov 20, 2026: ABOVE trend on 3 of 9 windows (20 years, 30 years, 40 years), BELOW on 6 (1 month, 3 months, 6 months, 1 year, 5 years, 10 years). Long-run (all-history) trend: 14.6% above.

    ●●●●●●○○○ Below trend on 6 of 9
    1 month
    -5.03%
    3 months
    -7.88%(weak trend)
    6 months
    -9.70%(weak trend)
    1 year
    -11.62%
    5 years
    -9.15%
    10 years
    -4.53%
    20 years
    +4.55%
    30 years
    +39.63%
    40 years
    +24.66%
    ◀ cheaper than trendpricier than trend ▶

    Price and trend lines, last 5 years

    $745.29 $295.19 Nov 2021 Nov 2026 1m 3m 6m 1y 5y 10y 20y 30y 40y all

    Solid line: price. Amber: the all-history trend (since Jan 1980, refitted every night) - how far the price is above or below its long-run path. Dashed: each trend line over its own window, extended to Nov 23; fainter = weaker fit (lower R²). Where the dashed lines bunch together the trends agree; where they fan out they disagree. VFINX price is dividend-adjusted (dividends reinvested) in today’s NAV dollars, the same basis as the trends, so older points sit below the NAV quoted on that day.

    Next trading day: Mon Nov 23, 2026. Source: Yahoo Finance, Vanguard 500 Index Fund Investor Shares VFINX (daily NAV since 1980); trends are fitted to dividend-adjusted prices (dividends reinvested), shown in today's NAV dollars. Made 2026-11-21 08:49 EST.

    WindowTrend todayGapAbove/Below On-trend price for Nov 23Trend growth/yrR²Bars
    1 month$672.48-5.03%BELOW$668.02-55.49%0.5522
    3 months$693.27-7.88%BELOW$692.49-12.83%0.2364
    6 months$707.25-9.70%BELOW$707.52+4.85%0.11126
    1 year$722.60-11.62%BELOW$723.59+18.09%0.72251
    5 years$702.99-9.15%BELOW$703.90+17.19%0.901,255
    10 years$668.93-4.53%BELOW$669.66+14.22%0.962,514
    20 years$610.85+4.55%ABOVE$611.44+12.40%0.955,031
    30 years$457.41+39.63%ABOVE$457.71+8.49%0.887,547
    40 years$512.31+24.66%ABOVE$512.70+9.55%0.9510,076
    All history (since Jan 1980)$557.47+14.56%ABOVE$557.92+10.19%0.9611,819

    How to read it: "On-trend price" is where the trend line for that window is on the next trading day. A NAV under it is BELOW that trend (cheaper than the trend), over it is ABOVE. R² near 1 = prices hugged the line; near 0 = no clear trend, so that row means little.

    Trend lines are fitted to past prices; they are not a forecast and not financial advice.

    VBMFX vs TRENDmade Sat Nov 21

    VBMFX vs trend - 2026-11-20

    VBMFX closed at $9.38 on Fri Nov 20, 2026: ABOVE trend on 6 of 8 windows (1 month, 3 months, 6 months, 1 year, 5 years, 10 years), BELOW on 2 (20 years, 30 years). Long-run (all-history) trend: 26.8% below.

    ○○○○○○●● Below trend on 2 of 8
    1 month
    +0.40%(weak trend)
    3 months
    +0.84%(weak trend)
    6 months
    +1.17%(weak trend)
    1 year
    +0.50%(weak trend)
    5 years
    +0.21%(weak trend)
    10 years
    +1.03%(weak trend)
    20 years
    -6.84%
    30 years
    -16.06%
    ◀ cheaper than trendpricier than trend ▶

    Price and trend lines, last 5 years

    $13.20 $7.70 Nov 2021 Nov 2026 1m 3m 6m 1y 5y 10y 20y 30y all

    Solid line: price. Amber: the all-history trend (since Dec 1986, refitted every night) - how far the price is above or below its long-run path. Dashed: each trend line over its own window, extended to Nov 23; fainter = weaker fit (lower R²). Where the dashed lines bunch together the trends agree; where they fan out they disagree. VBMFX price is dividend-adjusted (the monthly interest payments reinvested) in today’s NAV dollars, the same basis as the trends; most of a bond fund’s return is that interest, so older points sit far below the NAV quoted on that day.

    Next trading day: Mon Nov 23, 2026. Source: Yahoo Finance, Vanguard Total Bond Market Index Fund Investor Shares VBMFX (daily NAV since Dec 1986); trends are fitted to dividend-adjusted prices (monthly interest dividends reinvested), shown in today's NAV dollars. Made 2026-11-21 08:49 EST.

    WindowTrend todayGapAbove/Below On-trend price for Nov 23Trend growth/yrR²Bars
    1 month$9.34+0.40%ABOVE$9.34-3.75%0.0322
    3 months$9.30+0.84%ABOVE$9.30-1.21%0.0164
    6 months$9.27+1.17%ABOVE$9.27-4.68%0.48126
    1 year$9.33+0.50%ABOVE$9.33-1.87%0.36251
    5 years$9.36+0.21%ABOVE$9.36+2.12%0.391,255
    10 years$9.28+1.03%ABOVE$9.28+1.00%0.222,514
    20 years$10.07-6.84%BELOW$10.07+2.60%0.835,031
    30 years$11.18-16.06%BELOW$11.18+3.90%0.927,547
    All history (since Dec 1986)$12.81-26.78%BELOW$12.82+5.12%0.9410,063

    How to read it: "On-trend price" is where the trend line for that window is on the next trading day. A NAV under it is BELOW that trend (cheaper than the trend), over it is ABOVE. R² near 1 = prices hugged the line; near 0 = no clear trend, so that row means little.

    Trend lines are fitted to past prices; they are not a forecast and not financial advice.

    GOLD vs TRENDmade Sat Nov 21

    Gold vs trend - 2026-11-20

    Gold closed at $3,515.40 on Fri Nov 20, 2026: BELOW trend on 6 of 7 windows (1 month, 3 months, 6 months, 1 year, 5 years, 10 years). Long-run (all-history) trend: 4.2% above.

    ●●●●●●○ Below trend on 6 of 7
    1 month
    -7.13%(weak trend)
    3 months
    -8.22%
    6 months
    -13.86%(weak trend)
    1 year
    -14.75%(weak trend)
    5 years
    -23.23%
    10 years
    -4.45%
    20 years
    +25.70%
    ◀ cheaper than trendpricier than trend ▶

    Price and trend lines, last 5 years

    $5,478 $1,420 Nov 2021 Nov 2026 1m 3m 6m 1y 5y 10y 20y all

    Solid line: price. Amber: the all-history trend (since Aug 2000, refitted every night) - how far the price is above or below its long-run path. Dashed: each trend line over its own window, extended to Nov 23; fainter = weaker fit (lower R²). Where the dashed lines bunch together the trends agree; where they fan out they disagree. Gold price = front-month COMEX gold futures daily close (Yahoo GC=F), the same data the trends are fitted to.

    Next trading day: Mon Nov 23, 2026. Source: Yahoo Finance, COMEX gold futures GC=F (USD/oz, daily close). Made 2026-11-21 08:49 EST.

    WindowTrend todayGapAbove/Below On-trend price for Nov 23Trend growth/yrR²Bars
    1 month$3,785.44-7.13%BELOW$3,751.45-66.66%0.4822
    3 months$3,830.24-8.22%BELOW$3,806.42-53.21%0.8864
    6 months$4,080.93-13.86%BELOW$4,076.36-12.77%0.14126
    1 year$4,123.80-14.75%BELOW$4,118.61-14.21%0.31251
    5 years$4,578.85-23.23%BELOW$4,587.45+25.65%0.891,258
    10 years$3,678.97-4.45%BELOW$3,682.82+13.58%0.862,516
    20 years$2,796.60+25.70%ABOVE$2,798.03+6.40%0.705,031
    All history (since Aug 2000)$3,375.05+4.16%ABOVE$3,377.45+9.06%0.866,583

    How to read it: "On-trend price" is where the trend line for that window is on the next trading day. A price under it is BELOW that trend (cheaper than the trend), over it is ABOVE. R² near 1 = prices hugged the line; near 0 = no clear trend, so that row means little.

    Trend lines are fitted to past prices; they are not a forecast and not financial advice.

    HOW TO READ THIS PAGEwhat each part means and how it is made

    timodonnell.info/daily · guide

    Reading The-Good-Guy-Daily

    One web page, rebuilt every night on a small computer at home: a report on the US economy, then three markets measured against their own price trends: VFINX for stocks (an S&P 500 index fund), VBMFX for bonds (a total US bond market index fund) and gold. This guide explains how it is made and what every part is telling you.

    Plain words the idea, no finance background needed Precisely exact definitions, data and method
    What it isFour reports, one page

    US economy first, then stocks (VFINX), bonds (VBMFX) and gold, each measured against its trend under its own blue band. The links at the top jump to each one.

    Who writes itMostly scripts, a little AI

    Every number is fetched and calculated by ordinary programs. A small AI model running on the home computer writes only the short news notes, and those are fact-checked.

    WhenPublished around 3 a.m. Eastern

    The data is the latest available overnight: yesterday's closing prices and whatever official figures were released by then.

    How one night builds the page

    The computer wakes itself at 1:00 a.m., does about five minutes of real work spread over two hours, publishes, and goes back to sleep.

    1. Wake upsystem

      The computer wakes from sleep and is held awake until the night's work is done.

    2. Library downloadscript

      New official documents from the Federal Reserve, Bureau of Labor Statistics (BLS), Bureau of Economic Analysis (BEA) and Congressional Budget Office (CBO) are saved to a local library, along with their text.

    3. Collectscript + AI

      A script fetches about 20 economic series from FRED (the St. Louis Fed's database), the agencies' release feeds, AP and Reuters headlines, and a short reading pack of new official text. The AI model reads that pack and writes a few short news notes in a fixed format.

    4. Direction calls, history comparison, morning notescript

      Rules turn the latest readings into five direction calls, compare today's conditions with every past episode since the 1940s, and write the "What it means" note. No AI involved.

    5. Fact checkAI, checked

      Each news-note sentence is put back next to its source text and checked one at a time. Wrong sentences are replaced by the source's own sentence; unsupported ones are removed.

    6. Economy report builtscript

      A second, purely numeric check drops any sentence whose figure does not match the night's data.

    7. Gold, VFINX and VBMFX reportsscript

      Full daily price histories are downloaded from Yahoo Finance, the trend lines are fitted and the charts drawn. About one second each. Weekdays only.

    8. Publishedscript

      The three reports are joined into one page and pushed to the website. A backup build at 06:40 and a second publish at 06:50 catch a late or failed run.

    No night, no report.

    If the computer is off overnight, the missed night is not faked in the morning. The next real run looks back over the whole gap and marks the report as a catch-up. A report that is more than a day old (four days for the price reports, to cover weekends) is labelled "not updated today".

    Part one

    The US economy report

    A morning briefing built from official data. Read top to bottom, it goes from "what happened" to "what it might mean" to "how unusual is this, historically".

    Morning note

    Example, Nov 20, 2026: "Initial jobless claims (weekly): 203,000 for Nov 14, up from 202,000."

    Plain words

    The two-minute version. What changed lists the official numbers that came out since the last report. What it means puts them in context: are any recession warning signs on, and which way are things moving. Watch next shows the big releases coming up.

    A green tick under it means every number in the note was checked against the night's data.

    Precisely

    • "What changed" lists only observations whose release is newer than the previous report (not merely newer than the previous calendar day), with the prior value.
    • "What it means" is generated by script from the computed history pack: the warning-light count, the conditional recession frequency for months like this one against the unconditional base rate, and the five current direction calls.
    • "Watch next" is parsed from the agencies' official release calendars, with a cached fallback; the report warns if a calendar could not be read.

    Where it's heading (direction calls and scorecard)

    Example: "Inflation: up (high confidence) - CPI inflation y/y 3.72% for 2026-09 (was 3.35)."

    Plain words

    Five simple questions, answered every day: is growth, jobs, inflation, interest rates and recession risk going up, down or holding steady?

    Each answer is a vote of three related numbers, so one odd reading cannot swing it. "High confidence" means all three agree. Each call carries its own test ("wrong if…") and is graded later, so you can see over time whether the calls are any good.

    Precisely

    • Votes compare the latest reading with the previous one (a month earlier for daily series), with a dead band per series. Net +2 or more = rising, −2 or less = falling, otherwise steady; |3| = high confidence.
    • Growth: real GDP growth (SAAR), real retail sales, industrial production. Jobs: unemployment, initial claims, Sahm indicator (all inverted). Inflation: CPI, core CPI, core PCE y/y. Rates: fed funds target, 10-yr Treasury, 30-yr mortgage. Recession risk: Sahm, 10y–3m curve (inverted), Baa–10y spread.
    • A call stands until its direction changes or 35 days pass, then is scored RIGHT / WRONG / NO CHANGE on its key series. Results accumulate in the scorecard line.
    • These are descriptive momentum calls, not forecasts with a model behind them.

    Key numbers

    Example row: "Unemployment rate 4.1% · 2026-10 · ▼ 4.2 · 4.4"

    Plain words

    A dashboard of about 20 of the most-watched economic numbers: output, jobs, prices, interest rates, borrowing costs, the consumer mood, housing and oil.

    Each row shows the latest value, the date it covers, an arrow for which way it moved since the previous reading, and the value about a year ago. NEW marks numbers released since the last report.

    Precisely

    • Source: FRED series (BLS, BEA, Fed H.15, Freddie Mac PMMS, ICE BofA, Moody's, EIA via FRED); consumer sentiment from the University of Michigan's own page when it is newer than FRED.
    • Inflation, real retail sales and industrial production are y/y %; payrolls are the monthly change in thousands; GDP is the annualized quarterly real growth rate.
    • "Period" is the observation date, not the release date; the arrow compares with the prior observation (▲ up, ▼ down, ■ unchanged).
    • Spreads: high-yield = ICE BofA US HY OAS; Baa = Moody's Baa minus 10-yr Treasury.

    Today vs history, and the recession warning lights

    Example: "Recession warning lights: 0 of 8 on. Since 1986, in months with no lights on, a recession began within 12 months 2% of the time (4 of 241). Base rate for any month: 11%."

    Plain words

    This is the "has this happened before, and what came next?" section. It is pure arithmetic on long official records; no AI touches it.

    Eight warning lights are classic recession signals, such as rising unemployment or an inverted yield curve. The fewer that are on, the calmer the picture. Each one is compared with how often a recession followed in the past.

    Then every unusual condition that is true today (say, inflation above 3%) is checked against each past time it happened: did a recession follow, and did unemployment rise?

    Precisely

    • Recessions are NBER-dated (FRED USREC), monthly data. Lights: unemployment +0.5 pt over its 12-month low; 10y–3m inverted; 10y–2y inverted; claims 20%+ over 12-month low; Baa spread +1 pt over 12-month low; industrial production below a year ago; housing starts 15%+ below a year ago; 3-month average payroll growth under 50k.
    • The light panel uses the common sample of all eight (1986 on), excludes months inside recessions and the last 12 months, and reports the share of matching months followed by a recession start within 12 months, against the unconditional base rate.
    • Fifteen conditions in all, each with its own history (some back to 1948). An episode counts when the condition begins; the table gives episodes, recession within 12 and 24 months, and the typical change in unemployment 12 months later, against the base rate since that series starts.
    • Small samples (often 3–16 episodes) mean wide uncertainty: "0 of 3" is weak evidence either way.

    Official releases and headlines

    Example: "Fed 2026-11-18 — Minutes of the Federal Open Market Committee, October 27-28, 2026"

    Plain words

    Two plain lists with links: what the government agencies published in the last three days, and the last day's economic headlines from AP and Reuters. Nothing is summarised here; it is the raw "what came out" so you can click through.

    Precisely

    • Releases come from the Fed, BLS, BEA and CBO feeds plus the local document library (dated by release, so FOMC minutes appear about three weeks after the meeting).
    • AP and Reuters block scripts, so headlines come through Google News RSS. Titles only; their content is not used as data.

    News notes (the AI-written part)

    Example: "[DATA] Value added for outdoor recreation accounted for 2.3 percent ($704.2 billion) of current-dollar GDP in 2025."

    Plain words

    Short summaries of new official documents, written by the small AI model and then checked. Every note names its source, and every sentence starts with a tag that tells you what kind of statement it is:

    [DATA]a number taken from the source [REPORTED]what the source says happened [OPINION]someone's view, as stated in the source [ANALYSIS]an interpretation, not a fact

    A "corrected by fact check" mark means the AI's sentence was replaced with the source's own wording. "GAPS" lists what the source did not say.

    Precisely

    • Model: a ~2B-parameter Gemma model run locally on the CPU. It works mainly from the night's reading pack of official text; scripts do the fetching and all saving, and the model's reply is parsed from a fixed format.
    • Fact check, one sentence per prompt with the closest source passages beside it: replaced with the closest source sentence if judged wrong, removed if unsupported, kept only if judged right and every number in it appears in the source, "unchecked" if the source could not be fetched. Separately, a script check in the report build drops sentences that misread the night's data.
    • Treat these notes as pointers to the documents, not as analysis. The numbers that matter are in the tables above, which no model writes.

    History comparisons and open direction calls

    Example: "T-3: Inflation: rising — confidence high; wrong if: the next CPI inflation y/y reading is below 3.72"

    Plain words

    The record-keeping at the end. History comparisons (S-numbers) spell out in sentences what the history table showed. Open direction calls (T-numbers) list each current call with the exact result that would prove it wrong, so the scorecard can grade it honestly later.

    Precisely

    • Up to three conditions true today are written up, each with its episode count and 12/24-month recession frequencies.
    • "Wrong if" is fixed when the call is made (the key series and its baseline value), so a call cannot quietly change its own test.

    Part two

    VFINX, VBMFX and Gold vs trend

    Three reports built the same way, with no AI at all. They ask one question: is the price today above or below where its past trends say it "should" be, over many different time spans at once?

    How the chart works (illustration, not real data)

    5 years ago next trading day 5y 1y all 20y dot = latest close gap to the long-run trend

    Solid dark line: the price. Dashed lines: one trend per time span, fainter when the fit is weak. Solid amber line: the all-history trend. Lines bunching together means the trends agree; fanning out means they disagree.

    The trend idea

    Example, Nov 19, 2026: VFINX closed at $663.15; its all-history trend says $557.30, so the price is 19.0% above it.

    Plain words

    Prices grow by percentages, so over time they curve upward. For each time span (the last month, year, 10 years and so on), the report finds the steady growth rate that best fits the prices in that span. That is the trend line.

    If today's price is above a trend line, the price has run ahead of that trend; below it, the price is behind. Short spans show recent momentum; long spans show the big picture.

    A trend line describes the past. It is not a prediction, and the report says so.

    Precisely

    • Ordinary least squares of ln(price) on calendar days within each window: ln P = a + b·t. The fitted curve is exponential, a straight line on a log scale.
    • Windows in calendar days: 30, 91, 182, 365, 1826, 3652, 7305 (all three), plus 10957 (VFINX, VBMFX) and 14610 (VFINX only). A window with less than its full history (more than 7 days short) is flagged "short history".
    • Growth %/yr = e365.25·b − 1. R² is the fit of the log regression: near 1 the prices hugged the line, near 0 the window had no clear trend.
    • Each window is refitted from scratch every night. Fits are exact closed-form OLS, checked against an independent implementation to the cent.

    The data behind each report

    VFINX: 11,818 trading days since Jan 2, 1980. VBMFX: 10,062 since Dec 11, 1986. Gold: 6,582 since Aug 30, 2000.

    Plain words

    VFINX is Vanguard's 500 Index Fund, which tracks the S&P 500. Its prices include reinvested dividends, so they show what an investor actually earned.

    VBMFX is Vanguard's Total Bond Market Index Fund, the first bond index fund for ordinary investors (1986). It holds thousands of US government, mortgage and high-quality corporate bonds. Most of its return is the interest it pays each month, so its prices also include that interest reinvested. When interest rates rise, bond prices fall, which is why bonds dropped sharply in 2022.

    Gold is the price of gold futures on the COMEX exchange, in US dollars per ounce.

    Both use the previous trading day's closing price.

    Precisely

    • VFINX: Yahoo Finance daily history, adjusted close (dividends and distributions reinvested), rescaled so the latest adjusted close equals the latest NAV. Older points therefore sit below the NAV quoted at the time. Today's bar is ignored until 18:00 New York time, when the NAV is out.
    • VBMFX: the same method (Yahoo adjusted close, monthly income dividends reinvested, rescaled to the latest NAV), from the fund's first day, Dec 11, 1986, at a $10.00 NAV. The raw NAV has stayed near $10, so nearly all of the trend is reinvested income; the index it tracks today is the Bloomberg US Aggregate Float Adjusted.
    • Gold: Yahoo Finance GC=F, the continuous front-month COMEX contract, daily close; today's bar is ignored until 17:00 New York time. Roll effects between contracts are not adjusted.
    • "Next trading day" is the next weekday; exchange holidays are not modelled.

    Summary line, dots and gap bars

    Example: "BELOW trend on 6 of 9" with six red dots and three hollow green ones.

    Plain words

    The bold line at the top says how many trends the price is above and below. The row of dots says the same at a glance: red for below, green for above.

    Each bar shows how far the price is from one trend: left and red is cheaper than the trend, right and green is pricier. Pale rows marked "weak trend" are spans where prices did not follow a clear trend, so give them less weight.

    Precisely

    • Gap % = close / fitted value on the close date − 1, per window.
    • Bar length is linear in the gap and capped at ±20%; the printed number is exact.
    • "Weak trend" = R² below 0.5 (drawn at 40% opacity).
    • The all-history trend is reported alongside, not counted in the "N of 9" tally.

    Price and trend lines chart, with the amber line

    Example: on Nov 19, 2026 the gold chart showed the short-term trends ending above the price and the long ones below it.

    Plain words

    The last five years of price, with every trend line drawn faintly behind it, each running across its own time span and ending at the next trading day. The labels on the right name each line.

    When the lines bunch together, short and long trends agree. When they fan out, the market's recent path differs from its long-run path.

    The solid amber line is the trend through the whole price history, recalculated every night. The distance between the price and the amber line is the simplest answer to "how stretched is this market compared with its long-run path?"

    On a phone the chart opens at today; swipe left for earlier years.

    Precisely

    • Log price scale, so equal vertical distances are equal percentage changes and every exponential trend is a straight line. The y-range is the visible prices and line ends, padded 3%.
    • Price series thinned to at most 500 evenly spaced points for display, so a single-day extreme may not appear; the numbers in the tables use every day.
    • Line opacity = max(0.15, R²). Labels keep the order of the line ends and are pushed at least 10 px apart, so a label's height is approximate when lines converge.
    • Amber = the same OLS over the full sample (VFINX since Jan 1980, VBMFX since Dec 1986, gold since Aug 2000), refitted nightly: as of Nov 19, 2026, 10.2%/yr for VFINX (R² 0.96), 5.1%/yr for VBMFX (R² 0.94) and 9.1%/yr for gold (R² 0.86). A new day moves it by about 0.005%, so it drifts visibly only over months.

    The table

    Example row: "All history (since Jan 1980) · $557.30 · +18.99% · ABOVE · $557.45 · +10.19% · 0.96 · 11,818"

    Plain words

    ColumnMeans
    Trend todayWhere the trend line says the price would be today
    GapHow far the actual price is from that
    On-trend priceThe price that would sit exactly on the trend next trading day
    Growth/yrHow fast that trend climbs (or falls) per year
    R²How well the trend fits, 0 to 1
    BarsHow many trading days the trend is based on

    Precisely

    • Trend today = exp(a + b·t) at the last close date; on-trend price = the same at the next weekday.
    • All dollar figures are in today's NAV (VFINX) or today's contract (gold) terms.
    • Short-window growth rates are annualized from weeks of data and swing wildly (for example −47%/yr on a 1-month window); read them as slope, not expectation.
    • The amber-bordered last row is the all-history trend.

    What it is not

    Not financial advice

    The trends describe past prices. Markets can stay above or below any trend for years.

    Not a forecast model

    The direction calls and history tables count what happened before. They do not predict what happens next.

    Not real time

    Everything reflects the previous close and the official releases out by about 2 a.m. Eastern.

    Not written by AI where it counts

    Every number, table, chart and call is calculated by scripts. The AI writes only the tagged news notes, which are fact-checked twice.